nbd/client: Trace server noncompliance on structured reads
[qemu/ar7.git] / block / io.c
blobdfc153b8d8209d60237bca6ae1f60941ce407021
1 /*
2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
25 #include "qemu/osdep.h"
26 #include "trace.h"
27 #include "sysemu/block-backend.h"
28 #include "block/aio-wait.h"
29 #include "block/blockjob.h"
30 #include "block/blockjob_int.h"
31 #include "block/block_int.h"
32 #include "qemu/cutils.h"
33 #include "qapi/error.h"
34 #include "qemu/error-report.h"
36 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
38 /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
39 #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
41 static void bdrv_parent_cb_resize(BlockDriverState *bs);
42 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
43 int64_t offset, int bytes, BdrvRequestFlags flags);
45 void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore,
46 bool ignore_bds_parents)
48 BdrvChild *c, *next;
50 QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
51 if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) {
52 continue;
54 bdrv_parent_drained_begin_single(c, false);
58 void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore,
59 bool ignore_bds_parents)
61 BdrvChild *c, *next;
63 QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
64 if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) {
65 continue;
67 if (c->role->drained_end) {
68 c->role->drained_end(c);
73 static bool bdrv_parent_drained_poll_single(BdrvChild *c)
75 if (c->role->drained_poll) {
76 return c->role->drained_poll(c);
78 return false;
81 static bool bdrv_parent_drained_poll(BlockDriverState *bs, BdrvChild *ignore,
82 bool ignore_bds_parents)
84 BdrvChild *c, *next;
85 bool busy = false;
87 QLIST_FOREACH_SAFE(c, &bs->parents, next_parent, next) {
88 if (c == ignore || (ignore_bds_parents && c->role->parent_is_bds)) {
89 continue;
91 busy |= bdrv_parent_drained_poll_single(c);
94 return busy;
97 void bdrv_parent_drained_begin_single(BdrvChild *c, bool poll)
99 if (c->role->drained_begin) {
100 c->role->drained_begin(c);
102 if (poll) {
103 BDRV_POLL_WHILE(c->bs, bdrv_parent_drained_poll_single(c));
107 static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src)
109 dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer);
110 dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer);
111 dst->opt_mem_alignment = MAX(dst->opt_mem_alignment,
112 src->opt_mem_alignment);
113 dst->min_mem_alignment = MAX(dst->min_mem_alignment,
114 src->min_mem_alignment);
115 dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov);
118 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
120 BlockDriver *drv = bs->drv;
121 Error *local_err = NULL;
123 memset(&bs->bl, 0, sizeof(bs->bl));
125 if (!drv) {
126 return;
129 /* Default alignment based on whether driver has byte interface */
130 bs->bl.request_alignment = (drv->bdrv_co_preadv ||
131 drv->bdrv_aio_preadv) ? 1 : 512;
133 /* Take some limits from the children as a default */
134 if (bs->file) {
135 bdrv_refresh_limits(bs->file->bs, &local_err);
136 if (local_err) {
137 error_propagate(errp, local_err);
138 return;
140 bdrv_merge_limits(&bs->bl, &bs->file->bs->bl);
141 } else {
142 bs->bl.min_mem_alignment = 512;
143 bs->bl.opt_mem_alignment = getpagesize();
145 /* Safe default since most protocols use readv()/writev()/etc */
146 bs->bl.max_iov = IOV_MAX;
149 if (bs->backing) {
150 bdrv_refresh_limits(bs->backing->bs, &local_err);
151 if (local_err) {
152 error_propagate(errp, local_err);
153 return;
155 bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl);
158 /* Then let the driver override it */
159 if (drv->bdrv_refresh_limits) {
160 drv->bdrv_refresh_limits(bs, errp);
165 * The copy-on-read flag is actually a reference count so multiple users may
166 * use the feature without worrying about clobbering its previous state.
167 * Copy-on-read stays enabled until all users have called to disable it.
169 void bdrv_enable_copy_on_read(BlockDriverState *bs)
171 atomic_inc(&bs->copy_on_read);
174 void bdrv_disable_copy_on_read(BlockDriverState *bs)
176 int old = atomic_fetch_dec(&bs->copy_on_read);
177 assert(old >= 1);
180 typedef struct {
181 Coroutine *co;
182 BlockDriverState *bs;
183 bool done;
184 bool begin;
185 bool recursive;
186 bool poll;
187 BdrvChild *parent;
188 bool ignore_bds_parents;
189 } BdrvCoDrainData;
191 static void coroutine_fn bdrv_drain_invoke_entry(void *opaque)
193 BdrvCoDrainData *data = opaque;
194 BlockDriverState *bs = data->bs;
196 if (data->begin) {
197 bs->drv->bdrv_co_drain_begin(bs);
198 } else {
199 bs->drv->bdrv_co_drain_end(bs);
202 /* Set data->done before reading bs->wakeup. */
203 atomic_mb_set(&data->done, true);
204 bdrv_dec_in_flight(bs);
206 if (data->begin) {
207 g_free(data);
211 /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
212 static void bdrv_drain_invoke(BlockDriverState *bs, bool begin)
214 BdrvCoDrainData *data;
216 if (!bs->drv || (begin && !bs->drv->bdrv_co_drain_begin) ||
217 (!begin && !bs->drv->bdrv_co_drain_end)) {
218 return;
221 data = g_new(BdrvCoDrainData, 1);
222 *data = (BdrvCoDrainData) {
223 .bs = bs,
224 .done = false,
225 .begin = begin
228 /* Make sure the driver callback completes during the polling phase for
229 * drain_begin. */
230 bdrv_inc_in_flight(bs);
231 data->co = qemu_coroutine_create(bdrv_drain_invoke_entry, data);
232 aio_co_schedule(bdrv_get_aio_context(bs), data->co);
234 if (!begin) {
235 BDRV_POLL_WHILE(bs, !data->done);
236 g_free(data);
240 /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */
241 bool bdrv_drain_poll(BlockDriverState *bs, bool recursive,
242 BdrvChild *ignore_parent, bool ignore_bds_parents)
244 BdrvChild *child, *next;
246 if (bdrv_parent_drained_poll(bs, ignore_parent, ignore_bds_parents)) {
247 return true;
250 if (atomic_read(&bs->in_flight)) {
251 return true;
254 if (recursive) {
255 assert(!ignore_bds_parents);
256 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
257 if (bdrv_drain_poll(child->bs, recursive, child, false)) {
258 return true;
263 return false;
266 static bool bdrv_drain_poll_top_level(BlockDriverState *bs, bool recursive,
267 BdrvChild *ignore_parent)
269 return bdrv_drain_poll(bs, recursive, ignore_parent, false);
272 static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
273 BdrvChild *parent, bool ignore_bds_parents,
274 bool poll);
275 static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
276 BdrvChild *parent, bool ignore_bds_parents);
278 static void bdrv_co_drain_bh_cb(void *opaque)
280 BdrvCoDrainData *data = opaque;
281 Coroutine *co = data->co;
282 BlockDriverState *bs = data->bs;
284 if (bs) {
285 AioContext *ctx = bdrv_get_aio_context(bs);
286 AioContext *co_ctx = qemu_coroutine_get_aio_context(co);
289 * When the coroutine yielded, the lock for its home context was
290 * released, so we need to re-acquire it here. If it explicitly
291 * acquired a different context, the lock is still held and we don't
292 * want to lock it a second time (or AIO_WAIT_WHILE() would hang).
294 if (ctx == co_ctx) {
295 aio_context_acquire(ctx);
297 bdrv_dec_in_flight(bs);
298 if (data->begin) {
299 bdrv_do_drained_begin(bs, data->recursive, data->parent,
300 data->ignore_bds_parents, data->poll);
301 } else {
302 bdrv_do_drained_end(bs, data->recursive, data->parent,
303 data->ignore_bds_parents);
305 if (ctx == co_ctx) {
306 aio_context_release(ctx);
308 } else {
309 assert(data->begin);
310 bdrv_drain_all_begin();
313 data->done = true;
314 aio_co_wake(co);
317 static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs,
318 bool begin, bool recursive,
319 BdrvChild *parent,
320 bool ignore_bds_parents,
321 bool poll)
323 BdrvCoDrainData data;
325 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
326 * other coroutines run if they were queued by aio_co_enter(). */
328 assert(qemu_in_coroutine());
329 data = (BdrvCoDrainData) {
330 .co = qemu_coroutine_self(),
331 .bs = bs,
332 .done = false,
333 .begin = begin,
334 .recursive = recursive,
335 .parent = parent,
336 .ignore_bds_parents = ignore_bds_parents,
337 .poll = poll,
339 if (bs) {
340 bdrv_inc_in_flight(bs);
342 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs),
343 bdrv_co_drain_bh_cb, &data);
345 qemu_coroutine_yield();
346 /* If we are resumed from some other event (such as an aio completion or a
347 * timer callback), it is a bug in the caller that should be fixed. */
348 assert(data.done);
351 void bdrv_do_drained_begin_quiesce(BlockDriverState *bs,
352 BdrvChild *parent, bool ignore_bds_parents)
354 assert(!qemu_in_coroutine());
356 /* Stop things in parent-to-child order */
357 if (atomic_fetch_inc(&bs->quiesce_counter) == 0) {
358 aio_disable_external(bdrv_get_aio_context(bs));
361 bdrv_parent_drained_begin(bs, parent, ignore_bds_parents);
362 bdrv_drain_invoke(bs, true);
365 static void bdrv_do_drained_begin(BlockDriverState *bs, bool recursive,
366 BdrvChild *parent, bool ignore_bds_parents,
367 bool poll)
369 BdrvChild *child, *next;
371 if (qemu_in_coroutine()) {
372 bdrv_co_yield_to_drain(bs, true, recursive, parent, ignore_bds_parents,
373 poll);
374 return;
377 bdrv_do_drained_begin_quiesce(bs, parent, ignore_bds_parents);
379 if (recursive) {
380 assert(!ignore_bds_parents);
381 bs->recursive_quiesce_counter++;
382 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
383 bdrv_do_drained_begin(child->bs, true, child, ignore_bds_parents,
384 false);
389 * Wait for drained requests to finish.
391 * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The
392 * call is needed so things in this AioContext can make progress even
393 * though we don't return to the main AioContext loop - this automatically
394 * includes other nodes in the same AioContext and therefore all child
395 * nodes.
397 if (poll) {
398 assert(!ignore_bds_parents);
399 BDRV_POLL_WHILE(bs, bdrv_drain_poll_top_level(bs, recursive, parent));
403 void bdrv_drained_begin(BlockDriverState *bs)
405 bdrv_do_drained_begin(bs, false, NULL, false, true);
408 void bdrv_subtree_drained_begin(BlockDriverState *bs)
410 bdrv_do_drained_begin(bs, true, NULL, false, true);
413 static void bdrv_do_drained_end(BlockDriverState *bs, bool recursive,
414 BdrvChild *parent, bool ignore_bds_parents)
416 BdrvChild *child, *next;
417 int old_quiesce_counter;
419 if (qemu_in_coroutine()) {
420 bdrv_co_yield_to_drain(bs, false, recursive, parent, ignore_bds_parents,
421 false);
422 return;
424 assert(bs->quiesce_counter > 0);
425 old_quiesce_counter = atomic_fetch_dec(&bs->quiesce_counter);
427 /* Re-enable things in child-to-parent order */
428 bdrv_drain_invoke(bs, false);
429 bdrv_parent_drained_end(bs, parent, ignore_bds_parents);
430 if (old_quiesce_counter == 1) {
431 aio_enable_external(bdrv_get_aio_context(bs));
434 if (recursive) {
435 assert(!ignore_bds_parents);
436 bs->recursive_quiesce_counter--;
437 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
438 bdrv_do_drained_end(child->bs, true, child, ignore_bds_parents);
443 void bdrv_drained_end(BlockDriverState *bs)
445 bdrv_do_drained_end(bs, false, NULL, false);
448 void bdrv_subtree_drained_end(BlockDriverState *bs)
450 bdrv_do_drained_end(bs, true, NULL, false);
453 void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent)
455 int i;
457 for (i = 0; i < new_parent->recursive_quiesce_counter; i++) {
458 bdrv_do_drained_begin(child->bs, true, child, false, true);
462 void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent)
464 int i;
466 for (i = 0; i < old_parent->recursive_quiesce_counter; i++) {
467 bdrv_do_drained_end(child->bs, true, child, false);
472 * Wait for pending requests to complete on a single BlockDriverState subtree,
473 * and suspend block driver's internal I/O until next request arrives.
475 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
476 * AioContext.
478 void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
480 assert(qemu_in_coroutine());
481 bdrv_drained_begin(bs);
482 bdrv_drained_end(bs);
485 void bdrv_drain(BlockDriverState *bs)
487 bdrv_drained_begin(bs);
488 bdrv_drained_end(bs);
491 static void bdrv_drain_assert_idle(BlockDriverState *bs)
493 BdrvChild *child, *next;
495 assert(atomic_read(&bs->in_flight) == 0);
496 QLIST_FOREACH_SAFE(child, &bs->children, next, next) {
497 bdrv_drain_assert_idle(child->bs);
501 unsigned int bdrv_drain_all_count = 0;
503 static bool bdrv_drain_all_poll(void)
505 BlockDriverState *bs = NULL;
506 bool result = false;
508 /* bdrv_drain_poll() can't make changes to the graph and we are holding the
509 * main AioContext lock, so iterating bdrv_next_all_states() is safe. */
510 while ((bs = bdrv_next_all_states(bs))) {
511 AioContext *aio_context = bdrv_get_aio_context(bs);
512 aio_context_acquire(aio_context);
513 result |= bdrv_drain_poll(bs, false, NULL, true);
514 aio_context_release(aio_context);
517 return result;
521 * Wait for pending requests to complete across all BlockDriverStates
523 * This function does not flush data to disk, use bdrv_flush_all() for that
524 * after calling this function.
526 * This pauses all block jobs and disables external clients. It must
527 * be paired with bdrv_drain_all_end().
529 * NOTE: no new block jobs or BlockDriverStates can be created between
530 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
532 void bdrv_drain_all_begin(void)
534 BlockDriverState *bs = NULL;
536 if (qemu_in_coroutine()) {
537 bdrv_co_yield_to_drain(NULL, true, false, NULL, true, true);
538 return;
541 /* AIO_WAIT_WHILE() with a NULL context can only be called from the main
542 * loop AioContext, so make sure we're in the main context. */
543 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
544 assert(bdrv_drain_all_count < INT_MAX);
545 bdrv_drain_all_count++;
547 /* Quiesce all nodes, without polling in-flight requests yet. The graph
548 * cannot change during this loop. */
549 while ((bs = bdrv_next_all_states(bs))) {
550 AioContext *aio_context = bdrv_get_aio_context(bs);
552 aio_context_acquire(aio_context);
553 bdrv_do_drained_begin(bs, false, NULL, true, false);
554 aio_context_release(aio_context);
557 /* Now poll the in-flight requests */
558 AIO_WAIT_WHILE(NULL, bdrv_drain_all_poll());
560 while ((bs = bdrv_next_all_states(bs))) {
561 bdrv_drain_assert_idle(bs);
565 void bdrv_drain_all_end(void)
567 BlockDriverState *bs = NULL;
569 while ((bs = bdrv_next_all_states(bs))) {
570 AioContext *aio_context = bdrv_get_aio_context(bs);
572 aio_context_acquire(aio_context);
573 bdrv_do_drained_end(bs, false, NULL, true);
574 aio_context_release(aio_context);
577 assert(bdrv_drain_all_count > 0);
578 bdrv_drain_all_count--;
581 void bdrv_drain_all(void)
583 bdrv_drain_all_begin();
584 bdrv_drain_all_end();
588 * Remove an active request from the tracked requests list
590 * This function should be called when a tracked request is completing.
592 static void tracked_request_end(BdrvTrackedRequest *req)
594 if (req->serialising) {
595 atomic_dec(&req->bs->serialising_in_flight);
598 qemu_co_mutex_lock(&req->bs->reqs_lock);
599 QLIST_REMOVE(req, list);
600 qemu_co_queue_restart_all(&req->wait_queue);
601 qemu_co_mutex_unlock(&req->bs->reqs_lock);
605 * Add an active request to the tracked requests list
607 static void tracked_request_begin(BdrvTrackedRequest *req,
608 BlockDriverState *bs,
609 int64_t offset,
610 uint64_t bytes,
611 enum BdrvTrackedRequestType type)
613 assert(bytes <= INT64_MAX && offset <= INT64_MAX - bytes);
615 *req = (BdrvTrackedRequest){
616 .bs = bs,
617 .offset = offset,
618 .bytes = bytes,
619 .type = type,
620 .co = qemu_coroutine_self(),
621 .serialising = false,
622 .overlap_offset = offset,
623 .overlap_bytes = bytes,
626 qemu_co_queue_init(&req->wait_queue);
628 qemu_co_mutex_lock(&bs->reqs_lock);
629 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
630 qemu_co_mutex_unlock(&bs->reqs_lock);
633 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
635 int64_t overlap_offset = req->offset & ~(align - 1);
636 uint64_t overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
637 - overlap_offset;
639 if (!req->serialising) {
640 atomic_inc(&req->bs->serialising_in_flight);
641 req->serialising = true;
644 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
645 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
648 static bool is_request_serialising_and_aligned(BdrvTrackedRequest *req)
651 * If the request is serialising, overlap_offset and overlap_bytes are set,
652 * so we can check if the request is aligned. Otherwise, don't care and
653 * return false.
656 return req->serialising && (req->offset == req->overlap_offset) &&
657 (req->bytes == req->overlap_bytes);
661 * Round a region to cluster boundaries
663 void bdrv_round_to_clusters(BlockDriverState *bs,
664 int64_t offset, int64_t bytes,
665 int64_t *cluster_offset,
666 int64_t *cluster_bytes)
668 BlockDriverInfo bdi;
670 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
671 *cluster_offset = offset;
672 *cluster_bytes = bytes;
673 } else {
674 int64_t c = bdi.cluster_size;
675 *cluster_offset = QEMU_ALIGN_DOWN(offset, c);
676 *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c);
680 static int bdrv_get_cluster_size(BlockDriverState *bs)
682 BlockDriverInfo bdi;
683 int ret;
685 ret = bdrv_get_info(bs, &bdi);
686 if (ret < 0 || bdi.cluster_size == 0) {
687 return bs->bl.request_alignment;
688 } else {
689 return bdi.cluster_size;
693 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
694 int64_t offset, uint64_t bytes)
696 /* aaaa bbbb */
697 if (offset >= req->overlap_offset + req->overlap_bytes) {
698 return false;
700 /* bbbb aaaa */
701 if (req->overlap_offset >= offset + bytes) {
702 return false;
704 return true;
707 void bdrv_inc_in_flight(BlockDriverState *bs)
709 atomic_inc(&bs->in_flight);
712 void bdrv_wakeup(BlockDriverState *bs)
714 aio_wait_kick();
717 void bdrv_dec_in_flight(BlockDriverState *bs)
719 atomic_dec(&bs->in_flight);
720 bdrv_wakeup(bs);
723 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
725 BlockDriverState *bs = self->bs;
726 BdrvTrackedRequest *req;
727 bool retry;
728 bool waited = false;
730 if (!atomic_read(&bs->serialising_in_flight)) {
731 return false;
734 do {
735 retry = false;
736 qemu_co_mutex_lock(&bs->reqs_lock);
737 QLIST_FOREACH(req, &bs->tracked_requests, list) {
738 if (req == self || (!req->serialising && !self->serialising)) {
739 continue;
741 if (tracked_request_overlaps(req, self->overlap_offset,
742 self->overlap_bytes))
744 /* Hitting this means there was a reentrant request, for
745 * example, a block driver issuing nested requests. This must
746 * never happen since it means deadlock.
748 assert(qemu_coroutine_self() != req->co);
750 /* If the request is already (indirectly) waiting for us, or
751 * will wait for us as soon as it wakes up, then just go on
752 * (instead of producing a deadlock in the former case). */
753 if (!req->waiting_for) {
754 self->waiting_for = req;
755 qemu_co_queue_wait(&req->wait_queue, &bs->reqs_lock);
756 self->waiting_for = NULL;
757 retry = true;
758 waited = true;
759 break;
763 qemu_co_mutex_unlock(&bs->reqs_lock);
764 } while (retry);
766 return waited;
769 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
770 size_t size)
772 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
773 return -EIO;
776 if (!bdrv_is_inserted(bs)) {
777 return -ENOMEDIUM;
780 if (offset < 0) {
781 return -EIO;
784 return 0;
787 typedef struct RwCo {
788 BdrvChild *child;
789 int64_t offset;
790 QEMUIOVector *qiov;
791 bool is_write;
792 int ret;
793 BdrvRequestFlags flags;
794 } RwCo;
796 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
798 RwCo *rwco = opaque;
800 if (!rwco->is_write) {
801 rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset,
802 rwco->qiov->size, rwco->qiov,
803 rwco->flags);
804 } else {
805 rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset,
806 rwco->qiov->size, rwco->qiov,
807 rwco->flags);
809 aio_wait_kick();
813 * Process a vectored synchronous request using coroutines
815 static int bdrv_prwv_co(BdrvChild *child, int64_t offset,
816 QEMUIOVector *qiov, bool is_write,
817 BdrvRequestFlags flags)
819 Coroutine *co;
820 RwCo rwco = {
821 .child = child,
822 .offset = offset,
823 .qiov = qiov,
824 .is_write = is_write,
825 .ret = NOT_DONE,
826 .flags = flags,
829 if (qemu_in_coroutine()) {
830 /* Fast-path if already in coroutine context */
831 bdrv_rw_co_entry(&rwco);
832 } else {
833 co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco);
834 bdrv_coroutine_enter(child->bs, co);
835 BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE);
837 return rwco.ret;
841 * Process a synchronous request using coroutines
843 static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf,
844 int nb_sectors, bool is_write, BdrvRequestFlags flags)
846 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf,
847 nb_sectors * BDRV_SECTOR_SIZE);
849 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
850 return -EINVAL;
853 return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS,
854 &qiov, is_write, flags);
857 /* return < 0 if error. See bdrv_write() for the return codes */
858 int bdrv_read(BdrvChild *child, int64_t sector_num,
859 uint8_t *buf, int nb_sectors)
861 return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0);
864 /* Return < 0 if error. Important errors are:
865 -EIO generic I/O error (may happen for all errors)
866 -ENOMEDIUM No media inserted.
867 -EINVAL Invalid sector number or nb_sectors
868 -EACCES Trying to write a read-only device
870 int bdrv_write(BdrvChild *child, int64_t sector_num,
871 const uint8_t *buf, int nb_sectors)
873 return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
876 int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,
877 int bytes, BdrvRequestFlags flags)
879 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, NULL, bytes);
881 return bdrv_prwv_co(child, offset, &qiov, true,
882 BDRV_REQ_ZERO_WRITE | flags);
886 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
887 * The operation is sped up by checking the block status and only writing
888 * zeroes to the device if they currently do not return zeroes. Optional
889 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
890 * BDRV_REQ_FUA).
892 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
894 int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags)
896 int ret;
897 int64_t target_size, bytes, offset = 0;
898 BlockDriverState *bs = child->bs;
900 target_size = bdrv_getlength(bs);
901 if (target_size < 0) {
902 return target_size;
905 for (;;) {
906 bytes = MIN(target_size - offset, BDRV_REQUEST_MAX_BYTES);
907 if (bytes <= 0) {
908 return 0;
910 ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL);
911 if (ret < 0) {
912 return ret;
914 if (ret & BDRV_BLOCK_ZERO) {
915 offset += bytes;
916 continue;
918 ret = bdrv_pwrite_zeroes(child, offset, bytes, flags);
919 if (ret < 0) {
920 return ret;
922 offset += bytes;
926 int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
928 int ret;
930 ret = bdrv_prwv_co(child, offset, qiov, false, 0);
931 if (ret < 0) {
932 return ret;
935 return qiov->size;
938 int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
940 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
942 if (bytes < 0) {
943 return -EINVAL;
946 return bdrv_preadv(child, offset, &qiov);
949 int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
951 int ret;
953 ret = bdrv_prwv_co(child, offset, qiov, true, 0);
954 if (ret < 0) {
955 return ret;
958 return qiov->size;
961 int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes)
963 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
965 if (bytes < 0) {
966 return -EINVAL;
969 return bdrv_pwritev(child, offset, &qiov);
973 * Writes to the file and ensures that no writes are reordered across this
974 * request (acts as a barrier)
976 * Returns 0 on success, -errno in error cases.
978 int bdrv_pwrite_sync(BdrvChild *child, int64_t offset,
979 const void *buf, int count)
981 int ret;
983 ret = bdrv_pwrite(child, offset, buf, count);
984 if (ret < 0) {
985 return ret;
988 ret = bdrv_flush(child->bs);
989 if (ret < 0) {
990 return ret;
993 return 0;
996 typedef struct CoroutineIOCompletion {
997 Coroutine *coroutine;
998 int ret;
999 } CoroutineIOCompletion;
1001 static void bdrv_co_io_em_complete(void *opaque, int ret)
1003 CoroutineIOCompletion *co = opaque;
1005 co->ret = ret;
1006 aio_co_wake(co->coroutine);
1009 static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
1010 uint64_t offset, uint64_t bytes,
1011 QEMUIOVector *qiov, int flags)
1013 BlockDriver *drv = bs->drv;
1014 int64_t sector_num;
1015 unsigned int nb_sectors;
1017 assert(!(flags & ~BDRV_REQ_MASK));
1018 assert(!(flags & BDRV_REQ_NO_FALLBACK));
1020 if (!drv) {
1021 return -ENOMEDIUM;
1024 if (drv->bdrv_co_preadv) {
1025 return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags);
1028 if (drv->bdrv_aio_preadv) {
1029 BlockAIOCB *acb;
1030 CoroutineIOCompletion co = {
1031 .coroutine = qemu_coroutine_self(),
1034 acb = drv->bdrv_aio_preadv(bs, offset, bytes, qiov, flags,
1035 bdrv_co_io_em_complete, &co);
1036 if (acb == NULL) {
1037 return -EIO;
1038 } else {
1039 qemu_coroutine_yield();
1040 return co.ret;
1044 sector_num = offset >> BDRV_SECTOR_BITS;
1045 nb_sectors = bytes >> BDRV_SECTOR_BITS;
1047 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1048 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1049 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
1050 assert(drv->bdrv_co_readv);
1052 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
1055 static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs,
1056 uint64_t offset, uint64_t bytes,
1057 QEMUIOVector *qiov, int flags)
1059 BlockDriver *drv = bs->drv;
1060 int64_t sector_num;
1061 unsigned int nb_sectors;
1062 int ret;
1064 assert(!(flags & ~BDRV_REQ_MASK));
1065 assert(!(flags & BDRV_REQ_NO_FALLBACK));
1067 if (!drv) {
1068 return -ENOMEDIUM;
1071 if (drv->bdrv_co_pwritev) {
1072 ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov,
1073 flags & bs->supported_write_flags);
1074 flags &= ~bs->supported_write_flags;
1075 goto emulate_flags;
1078 if (drv->bdrv_aio_pwritev) {
1079 BlockAIOCB *acb;
1080 CoroutineIOCompletion co = {
1081 .coroutine = qemu_coroutine_self(),
1084 acb = drv->bdrv_aio_pwritev(bs, offset, bytes, qiov,
1085 flags & bs->supported_write_flags,
1086 bdrv_co_io_em_complete, &co);
1087 flags &= ~bs->supported_write_flags;
1088 if (acb == NULL) {
1089 ret = -EIO;
1090 } else {
1091 qemu_coroutine_yield();
1092 ret = co.ret;
1094 goto emulate_flags;
1097 sector_num = offset >> BDRV_SECTOR_BITS;
1098 nb_sectors = bytes >> BDRV_SECTOR_BITS;
1100 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1101 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1102 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
1104 assert(drv->bdrv_co_writev);
1105 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov,
1106 flags & bs->supported_write_flags);
1107 flags &= ~bs->supported_write_flags;
1109 emulate_flags:
1110 if (ret == 0 && (flags & BDRV_REQ_FUA)) {
1111 ret = bdrv_co_flush(bs);
1114 return ret;
1117 static int coroutine_fn
1118 bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
1119 uint64_t bytes, QEMUIOVector *qiov)
1121 BlockDriver *drv = bs->drv;
1123 if (!drv) {
1124 return -ENOMEDIUM;
1127 if (!drv->bdrv_co_pwritev_compressed) {
1128 return -ENOTSUP;
1131 return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov);
1134 static int coroutine_fn bdrv_co_do_copy_on_readv(BdrvChild *child,
1135 int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
1137 BlockDriverState *bs = child->bs;
1139 /* Perform I/O through a temporary buffer so that users who scribble over
1140 * their read buffer while the operation is in progress do not end up
1141 * modifying the image file. This is critical for zero-copy guest I/O
1142 * where anything might happen inside guest memory.
1144 void *bounce_buffer;
1146 BlockDriver *drv = bs->drv;
1147 QEMUIOVector local_qiov;
1148 int64_t cluster_offset;
1149 int64_t cluster_bytes;
1150 size_t skip_bytes;
1151 int ret;
1152 int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer,
1153 BDRV_REQUEST_MAX_BYTES);
1154 unsigned int progress = 0;
1156 if (!drv) {
1157 return -ENOMEDIUM;
1160 /* FIXME We cannot require callers to have write permissions when all they
1161 * are doing is a read request. If we did things right, write permissions
1162 * would be obtained anyway, but internally by the copy-on-read code. As
1163 * long as it is implemented here rather than in a separate filter driver,
1164 * the copy-on-read code doesn't have its own BdrvChild, however, for which
1165 * it could request permissions. Therefore we have to bypass the permission
1166 * system for the moment. */
1167 // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1169 /* Cover entire cluster so no additional backing file I/O is required when
1170 * allocating cluster in the image file. Note that this value may exceed
1171 * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which
1172 * is one reason we loop rather than doing it all at once.
1174 bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
1175 skip_bytes = offset - cluster_offset;
1177 trace_bdrv_co_do_copy_on_readv(bs, offset, bytes,
1178 cluster_offset, cluster_bytes);
1180 bounce_buffer = qemu_try_blockalign(bs,
1181 MIN(MIN(max_transfer, cluster_bytes),
1182 MAX_BOUNCE_BUFFER));
1183 if (bounce_buffer == NULL) {
1184 ret = -ENOMEM;
1185 goto err;
1188 while (cluster_bytes) {
1189 int64_t pnum;
1191 ret = bdrv_is_allocated(bs, cluster_offset,
1192 MIN(cluster_bytes, max_transfer), &pnum);
1193 if (ret < 0) {
1194 /* Safe to treat errors in querying allocation as if
1195 * unallocated; we'll probably fail again soon on the
1196 * read, but at least that will set a decent errno.
1198 pnum = MIN(cluster_bytes, max_transfer);
1201 /* Stop at EOF if the image ends in the middle of the cluster */
1202 if (ret == 0 && pnum == 0) {
1203 assert(progress >= bytes);
1204 break;
1207 assert(skip_bytes < pnum);
1209 if (ret <= 0) {
1210 /* Must copy-on-read; use the bounce buffer */
1211 pnum = MIN(pnum, MAX_BOUNCE_BUFFER);
1212 qemu_iovec_init_buf(&local_qiov, bounce_buffer, pnum);
1214 ret = bdrv_driver_preadv(bs, cluster_offset, pnum,
1215 &local_qiov, 0);
1216 if (ret < 0) {
1217 goto err;
1220 bdrv_debug_event(bs, BLKDBG_COR_WRITE);
1221 if (drv->bdrv_co_pwrite_zeroes &&
1222 buffer_is_zero(bounce_buffer, pnum)) {
1223 /* FIXME: Should we (perhaps conditionally) be setting
1224 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
1225 * that still correctly reads as zero? */
1226 ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, pnum,
1227 BDRV_REQ_WRITE_UNCHANGED);
1228 } else {
1229 /* This does not change the data on the disk, it is not
1230 * necessary to flush even in cache=writethrough mode.
1232 ret = bdrv_driver_pwritev(bs, cluster_offset, pnum,
1233 &local_qiov,
1234 BDRV_REQ_WRITE_UNCHANGED);
1237 if (ret < 0) {
1238 /* It might be okay to ignore write errors for guest
1239 * requests. If this is a deliberate copy-on-read
1240 * then we don't want to ignore the error. Simply
1241 * report it in all cases.
1243 goto err;
1246 qemu_iovec_from_buf(qiov, progress, bounce_buffer + skip_bytes,
1247 pnum - skip_bytes);
1248 } else {
1249 /* Read directly into the destination */
1250 qemu_iovec_init(&local_qiov, qiov->niov);
1251 qemu_iovec_concat(&local_qiov, qiov, progress, pnum - skip_bytes);
1252 ret = bdrv_driver_preadv(bs, offset + progress, local_qiov.size,
1253 &local_qiov, 0);
1254 qemu_iovec_destroy(&local_qiov);
1255 if (ret < 0) {
1256 goto err;
1260 cluster_offset += pnum;
1261 cluster_bytes -= pnum;
1262 progress += pnum - skip_bytes;
1263 skip_bytes = 0;
1265 ret = 0;
1267 err:
1268 qemu_vfree(bounce_buffer);
1269 return ret;
1273 * Forwards an already correctly aligned request to the BlockDriver. This
1274 * handles copy on read, zeroing after EOF, and fragmentation of large
1275 * reads; any other features must be implemented by the caller.
1277 static int coroutine_fn bdrv_aligned_preadv(BdrvChild *child,
1278 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1279 int64_t align, QEMUIOVector *qiov, int flags)
1281 BlockDriverState *bs = child->bs;
1282 int64_t total_bytes, max_bytes;
1283 int ret = 0;
1284 uint64_t bytes_remaining = bytes;
1285 int max_transfer;
1287 assert(is_power_of_2(align));
1288 assert((offset & (align - 1)) == 0);
1289 assert((bytes & (align - 1)) == 0);
1290 assert(!qiov || bytes == qiov->size);
1291 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1292 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1293 align);
1295 /* TODO: We would need a per-BDS .supported_read_flags and
1296 * potential fallback support, if we ever implement any read flags
1297 * to pass through to drivers. For now, there aren't any
1298 * passthrough flags. */
1299 assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ)));
1301 /* Handle Copy on Read and associated serialisation */
1302 if (flags & BDRV_REQ_COPY_ON_READ) {
1303 /* If we touch the same cluster it counts as an overlap. This
1304 * guarantees that allocating writes will be serialized and not race
1305 * with each other for the same cluster. For example, in copy-on-read
1306 * it ensures that the CoR read and write operations are atomic and
1307 * guest writes cannot interleave between them. */
1308 mark_request_serialising(req, bdrv_get_cluster_size(bs));
1311 /* BDRV_REQ_SERIALISING is only for write operation */
1312 assert(!(flags & BDRV_REQ_SERIALISING));
1314 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
1315 wait_serialising_requests(req);
1318 if (flags & BDRV_REQ_COPY_ON_READ) {
1319 int64_t pnum;
1321 ret = bdrv_is_allocated(bs, offset, bytes, &pnum);
1322 if (ret < 0) {
1323 goto out;
1326 if (!ret || pnum != bytes) {
1327 ret = bdrv_co_do_copy_on_readv(child, offset, bytes, qiov);
1328 goto out;
1332 /* Forward the request to the BlockDriver, possibly fragmenting it */
1333 total_bytes = bdrv_getlength(bs);
1334 if (total_bytes < 0) {
1335 ret = total_bytes;
1336 goto out;
1339 max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align);
1340 if (bytes <= max_bytes && bytes <= max_transfer) {
1341 ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
1342 goto out;
1345 while (bytes_remaining) {
1346 int num;
1348 if (max_bytes) {
1349 QEMUIOVector local_qiov;
1351 num = MIN(bytes_remaining, MIN(max_bytes, max_transfer));
1352 assert(num);
1353 qemu_iovec_init(&local_qiov, qiov->niov);
1354 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1356 ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining,
1357 num, &local_qiov, 0);
1358 max_bytes -= num;
1359 qemu_iovec_destroy(&local_qiov);
1360 } else {
1361 num = bytes_remaining;
1362 ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0,
1363 bytes_remaining);
1365 if (ret < 0) {
1366 goto out;
1368 bytes_remaining -= num;
1371 out:
1372 return ret < 0 ? ret : 0;
1376 * Handle a read request in coroutine context
1378 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
1379 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1380 BdrvRequestFlags flags)
1382 BlockDriverState *bs = child->bs;
1383 BlockDriver *drv = bs->drv;
1384 BdrvTrackedRequest req;
1386 uint64_t align = bs->bl.request_alignment;
1387 uint8_t *head_buf = NULL;
1388 uint8_t *tail_buf = NULL;
1389 QEMUIOVector local_qiov;
1390 bool use_local_qiov = false;
1391 int ret;
1393 trace_bdrv_co_preadv(child->bs, offset, bytes, flags);
1395 if (!drv) {
1396 return -ENOMEDIUM;
1399 ret = bdrv_check_byte_request(bs, offset, bytes);
1400 if (ret < 0) {
1401 return ret;
1404 bdrv_inc_in_flight(bs);
1406 /* Don't do copy-on-read if we read data before write operation */
1407 if (atomic_read(&bs->copy_on_read) && !(flags & BDRV_REQ_NO_SERIALISING)) {
1408 flags |= BDRV_REQ_COPY_ON_READ;
1411 /* Align read if necessary by padding qiov */
1412 if (offset & (align - 1)) {
1413 head_buf = qemu_blockalign(bs, align);
1414 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1415 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1416 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1417 use_local_qiov = true;
1419 bytes += offset & (align - 1);
1420 offset = offset & ~(align - 1);
1423 if ((offset + bytes) & (align - 1)) {
1424 if (!use_local_qiov) {
1425 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1426 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1427 use_local_qiov = true;
1429 tail_buf = qemu_blockalign(bs, align);
1430 qemu_iovec_add(&local_qiov, tail_buf,
1431 align - ((offset + bytes) & (align - 1)));
1433 bytes = ROUND_UP(bytes, align);
1436 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
1437 ret = bdrv_aligned_preadv(child, &req, offset, bytes, align,
1438 use_local_qiov ? &local_qiov : qiov,
1439 flags);
1440 tracked_request_end(&req);
1441 bdrv_dec_in_flight(bs);
1443 if (use_local_qiov) {
1444 qemu_iovec_destroy(&local_qiov);
1445 qemu_vfree(head_buf);
1446 qemu_vfree(tail_buf);
1449 return ret;
1452 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
1453 int64_t offset, int bytes, BdrvRequestFlags flags)
1455 BlockDriver *drv = bs->drv;
1456 QEMUIOVector qiov;
1457 void *buf = NULL;
1458 int ret = 0;
1459 bool need_flush = false;
1460 int head = 0;
1461 int tail = 0;
1463 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX);
1464 int alignment = MAX(bs->bl.pwrite_zeroes_alignment,
1465 bs->bl.request_alignment);
1466 int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer, MAX_BOUNCE_BUFFER);
1468 if (!drv) {
1469 return -ENOMEDIUM;
1472 if ((flags & ~bs->supported_zero_flags) & BDRV_REQ_NO_FALLBACK) {
1473 return -ENOTSUP;
1476 assert(alignment % bs->bl.request_alignment == 0);
1477 head = offset % alignment;
1478 tail = (offset + bytes) % alignment;
1479 max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment);
1480 assert(max_write_zeroes >= bs->bl.request_alignment);
1482 while (bytes > 0 && !ret) {
1483 int num = bytes;
1485 /* Align request. Block drivers can expect the "bulk" of the request
1486 * to be aligned, and that unaligned requests do not cross cluster
1487 * boundaries.
1489 if (head) {
1490 /* Make a small request up to the first aligned sector. For
1491 * convenience, limit this request to max_transfer even if
1492 * we don't need to fall back to writes. */
1493 num = MIN(MIN(bytes, max_transfer), alignment - head);
1494 head = (head + num) % alignment;
1495 assert(num < max_write_zeroes);
1496 } else if (tail && num > alignment) {
1497 /* Shorten the request to the last aligned sector. */
1498 num -= tail;
1501 /* limit request size */
1502 if (num > max_write_zeroes) {
1503 num = max_write_zeroes;
1506 ret = -ENOTSUP;
1507 /* First try the efficient write zeroes operation */
1508 if (drv->bdrv_co_pwrite_zeroes) {
1509 ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num,
1510 flags & bs->supported_zero_flags);
1511 if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) &&
1512 !(bs->supported_zero_flags & BDRV_REQ_FUA)) {
1513 need_flush = true;
1515 } else {
1516 assert(!bs->supported_zero_flags);
1519 if (ret == -ENOTSUP && !(flags & BDRV_REQ_NO_FALLBACK)) {
1520 /* Fall back to bounce buffer if write zeroes is unsupported */
1521 BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
1523 if ((flags & BDRV_REQ_FUA) &&
1524 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1525 /* No need for bdrv_driver_pwrite() to do a fallback
1526 * flush on each chunk; use just one at the end */
1527 write_flags &= ~BDRV_REQ_FUA;
1528 need_flush = true;
1530 num = MIN(num, max_transfer);
1531 if (buf == NULL) {
1532 buf = qemu_try_blockalign0(bs, num);
1533 if (buf == NULL) {
1534 ret = -ENOMEM;
1535 goto fail;
1538 qemu_iovec_init_buf(&qiov, buf, num);
1540 ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags);
1542 /* Keep bounce buffer around if it is big enough for all
1543 * all future requests.
1545 if (num < max_transfer) {
1546 qemu_vfree(buf);
1547 buf = NULL;
1551 offset += num;
1552 bytes -= num;
1555 fail:
1556 if (ret == 0 && need_flush) {
1557 ret = bdrv_co_flush(bs);
1559 qemu_vfree(buf);
1560 return ret;
1563 static inline int coroutine_fn
1564 bdrv_co_write_req_prepare(BdrvChild *child, int64_t offset, uint64_t bytes,
1565 BdrvTrackedRequest *req, int flags)
1567 BlockDriverState *bs = child->bs;
1568 bool waited;
1569 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1571 if (bs->read_only) {
1572 return -EPERM;
1575 /* BDRV_REQ_NO_SERIALISING is only for read operation */
1576 assert(!(flags & BDRV_REQ_NO_SERIALISING));
1577 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1578 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1579 assert(!(flags & ~BDRV_REQ_MASK));
1581 if (flags & BDRV_REQ_SERIALISING) {
1582 mark_request_serialising(req, bdrv_get_cluster_size(bs));
1585 waited = wait_serialising_requests(req);
1587 assert(!waited || !req->serialising ||
1588 is_request_serialising_and_aligned(req));
1589 assert(req->overlap_offset <= offset);
1590 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1591 assert(end_sector <= bs->total_sectors || child->perm & BLK_PERM_RESIZE);
1593 switch (req->type) {
1594 case BDRV_TRACKED_WRITE:
1595 case BDRV_TRACKED_DISCARD:
1596 if (flags & BDRV_REQ_WRITE_UNCHANGED) {
1597 assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1598 } else {
1599 assert(child->perm & BLK_PERM_WRITE);
1601 return notifier_with_return_list_notify(&bs->before_write_notifiers,
1602 req);
1603 case BDRV_TRACKED_TRUNCATE:
1604 assert(child->perm & BLK_PERM_RESIZE);
1605 return 0;
1606 default:
1607 abort();
1611 static inline void coroutine_fn
1612 bdrv_co_write_req_finish(BdrvChild *child, int64_t offset, uint64_t bytes,
1613 BdrvTrackedRequest *req, int ret)
1615 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1616 BlockDriverState *bs = child->bs;
1618 atomic_inc(&bs->write_gen);
1621 * Discard cannot extend the image, but in error handling cases, such as
1622 * when reverting a qcow2 cluster allocation, the discarded range can pass
1623 * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD
1624 * here. Instead, just skip it, since semantically a discard request
1625 * beyond EOF cannot expand the image anyway.
1627 if (ret == 0 &&
1628 (req->type == BDRV_TRACKED_TRUNCATE ||
1629 end_sector > bs->total_sectors) &&
1630 req->type != BDRV_TRACKED_DISCARD) {
1631 bs->total_sectors = end_sector;
1632 bdrv_parent_cb_resize(bs);
1633 bdrv_dirty_bitmap_truncate(bs, end_sector << BDRV_SECTOR_BITS);
1635 if (req->bytes) {
1636 switch (req->type) {
1637 case BDRV_TRACKED_WRITE:
1638 stat64_max(&bs->wr_highest_offset, offset + bytes);
1639 /* fall through, to set dirty bits */
1640 case BDRV_TRACKED_DISCARD:
1641 bdrv_set_dirty(bs, offset, bytes);
1642 break;
1643 default:
1644 break;
1650 * Forwards an already correctly aligned write request to the BlockDriver,
1651 * after possibly fragmenting it.
1653 static int coroutine_fn bdrv_aligned_pwritev(BdrvChild *child,
1654 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1655 int64_t align, QEMUIOVector *qiov, int flags)
1657 BlockDriverState *bs = child->bs;
1658 BlockDriver *drv = bs->drv;
1659 int ret;
1661 uint64_t bytes_remaining = bytes;
1662 int max_transfer;
1664 if (!drv) {
1665 return -ENOMEDIUM;
1668 if (bdrv_has_readonly_bitmaps(bs)) {
1669 return -EPERM;
1672 assert(is_power_of_2(align));
1673 assert((offset & (align - 1)) == 0);
1674 assert((bytes & (align - 1)) == 0);
1675 assert(!qiov || bytes == qiov->size);
1676 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1677 align);
1679 ret = bdrv_co_write_req_prepare(child, offset, bytes, req, flags);
1681 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1682 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes &&
1683 qemu_iovec_is_zero(qiov)) {
1684 flags |= BDRV_REQ_ZERO_WRITE;
1685 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1686 flags |= BDRV_REQ_MAY_UNMAP;
1690 if (ret < 0) {
1691 /* Do nothing, write notifier decided to fail this request */
1692 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1693 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1694 ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags);
1695 } else if (flags & BDRV_REQ_WRITE_COMPRESSED) {
1696 ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov);
1697 } else if (bytes <= max_transfer) {
1698 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1699 ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags);
1700 } else {
1701 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1702 while (bytes_remaining) {
1703 int num = MIN(bytes_remaining, max_transfer);
1704 QEMUIOVector local_qiov;
1705 int local_flags = flags;
1707 assert(num);
1708 if (num < bytes_remaining && (flags & BDRV_REQ_FUA) &&
1709 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1710 /* If FUA is going to be emulated by flush, we only
1711 * need to flush on the last iteration */
1712 local_flags &= ~BDRV_REQ_FUA;
1714 qemu_iovec_init(&local_qiov, qiov->niov);
1715 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1717 ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining,
1718 num, &local_qiov, local_flags);
1719 qemu_iovec_destroy(&local_qiov);
1720 if (ret < 0) {
1721 break;
1723 bytes_remaining -= num;
1726 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1728 if (ret >= 0) {
1729 ret = 0;
1731 bdrv_co_write_req_finish(child, offset, bytes, req, ret);
1733 return ret;
1736 static int coroutine_fn bdrv_co_do_zero_pwritev(BdrvChild *child,
1737 int64_t offset,
1738 unsigned int bytes,
1739 BdrvRequestFlags flags,
1740 BdrvTrackedRequest *req)
1742 BlockDriverState *bs = child->bs;
1743 uint8_t *buf = NULL;
1744 QEMUIOVector local_qiov;
1745 uint64_t align = bs->bl.request_alignment;
1746 unsigned int head_padding_bytes, tail_padding_bytes;
1747 int ret = 0;
1749 head_padding_bytes = offset & (align - 1);
1750 tail_padding_bytes = (align - (offset + bytes)) & (align - 1);
1753 assert(flags & BDRV_REQ_ZERO_WRITE);
1754 if (head_padding_bytes || tail_padding_bytes) {
1755 buf = qemu_blockalign(bs, align);
1756 qemu_iovec_init_buf(&local_qiov, buf, align);
1758 if (head_padding_bytes) {
1759 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1761 /* RMW the unaligned part before head. */
1762 mark_request_serialising(req, align);
1763 wait_serialising_requests(req);
1764 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1765 ret = bdrv_aligned_preadv(child, req, offset & ~(align - 1), align,
1766 align, &local_qiov, 0);
1767 if (ret < 0) {
1768 goto fail;
1770 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1772 memset(buf + head_padding_bytes, 0, zero_bytes);
1773 ret = bdrv_aligned_pwritev(child, req, offset & ~(align - 1), align,
1774 align, &local_qiov,
1775 flags & ~BDRV_REQ_ZERO_WRITE);
1776 if (ret < 0) {
1777 goto fail;
1779 offset += zero_bytes;
1780 bytes -= zero_bytes;
1783 assert(!bytes || (offset & (align - 1)) == 0);
1784 if (bytes >= align) {
1785 /* Write the aligned part in the middle. */
1786 uint64_t aligned_bytes = bytes & ~(align - 1);
1787 ret = bdrv_aligned_pwritev(child, req, offset, aligned_bytes, align,
1788 NULL, flags);
1789 if (ret < 0) {
1790 goto fail;
1792 bytes -= aligned_bytes;
1793 offset += aligned_bytes;
1796 assert(!bytes || (offset & (align - 1)) == 0);
1797 if (bytes) {
1798 assert(align == tail_padding_bytes + bytes);
1799 /* RMW the unaligned part after tail. */
1800 mark_request_serialising(req, align);
1801 wait_serialising_requests(req);
1802 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1803 ret = bdrv_aligned_preadv(child, req, offset, align,
1804 align, &local_qiov, 0);
1805 if (ret < 0) {
1806 goto fail;
1808 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1810 memset(buf, 0, bytes);
1811 ret = bdrv_aligned_pwritev(child, req, offset, align, align,
1812 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1814 fail:
1815 qemu_vfree(buf);
1816 return ret;
1821 * Handle a write request in coroutine context
1823 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
1824 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1825 BdrvRequestFlags flags)
1827 BlockDriverState *bs = child->bs;
1828 BdrvTrackedRequest req;
1829 uint64_t align = bs->bl.request_alignment;
1830 uint8_t *head_buf = NULL;
1831 uint8_t *tail_buf = NULL;
1832 QEMUIOVector local_qiov;
1833 bool use_local_qiov = false;
1834 int ret;
1836 trace_bdrv_co_pwritev(child->bs, offset, bytes, flags);
1838 if (!bs->drv) {
1839 return -ENOMEDIUM;
1842 ret = bdrv_check_byte_request(bs, offset, bytes);
1843 if (ret < 0) {
1844 return ret;
1847 bdrv_inc_in_flight(bs);
1849 * Align write if necessary by performing a read-modify-write cycle.
1850 * Pad qiov with the read parts and be sure to have a tracked request not
1851 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1853 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1855 if (flags & BDRV_REQ_ZERO_WRITE) {
1856 ret = bdrv_co_do_zero_pwritev(child, offset, bytes, flags, &req);
1857 goto out;
1860 if (offset & (align - 1)) {
1861 QEMUIOVector head_qiov;
1863 mark_request_serialising(&req, align);
1864 wait_serialising_requests(&req);
1866 head_buf = qemu_blockalign(bs, align);
1867 qemu_iovec_init_buf(&head_qiov, head_buf, align);
1869 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1870 ret = bdrv_aligned_preadv(child, &req, offset & ~(align - 1), align,
1871 align, &head_qiov, 0);
1872 if (ret < 0) {
1873 goto fail;
1875 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1877 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1878 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1879 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1880 use_local_qiov = true;
1882 bytes += offset & (align - 1);
1883 offset = offset & ~(align - 1);
1885 /* We have read the tail already if the request is smaller
1886 * than one aligned block.
1888 if (bytes < align) {
1889 qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes);
1890 bytes = align;
1894 if ((offset + bytes) & (align - 1)) {
1895 QEMUIOVector tail_qiov;
1896 size_t tail_bytes;
1897 bool waited;
1899 mark_request_serialising(&req, align);
1900 waited = wait_serialising_requests(&req);
1901 assert(!waited || !use_local_qiov);
1903 tail_buf = qemu_blockalign(bs, align);
1904 qemu_iovec_init_buf(&tail_qiov, tail_buf, align);
1906 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1907 ret = bdrv_aligned_preadv(child, &req, (offset + bytes) & ~(align - 1),
1908 align, align, &tail_qiov, 0);
1909 if (ret < 0) {
1910 goto fail;
1912 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1914 if (!use_local_qiov) {
1915 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1916 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1917 use_local_qiov = true;
1920 tail_bytes = (offset + bytes) & (align - 1);
1921 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1923 bytes = ROUND_UP(bytes, align);
1926 ret = bdrv_aligned_pwritev(child, &req, offset, bytes, align,
1927 use_local_qiov ? &local_qiov : qiov,
1928 flags);
1930 fail:
1932 if (use_local_qiov) {
1933 qemu_iovec_destroy(&local_qiov);
1935 qemu_vfree(head_buf);
1936 qemu_vfree(tail_buf);
1937 out:
1938 tracked_request_end(&req);
1939 bdrv_dec_in_flight(bs);
1940 return ret;
1943 int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
1944 int bytes, BdrvRequestFlags flags)
1946 trace_bdrv_co_pwrite_zeroes(child->bs, offset, bytes, flags);
1948 if (!(child->bs->open_flags & BDRV_O_UNMAP)) {
1949 flags &= ~BDRV_REQ_MAY_UNMAP;
1952 return bdrv_co_pwritev(child, offset, bytes, NULL,
1953 BDRV_REQ_ZERO_WRITE | flags);
1957 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
1959 int bdrv_flush_all(void)
1961 BdrvNextIterator it;
1962 BlockDriverState *bs = NULL;
1963 int result = 0;
1965 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
1966 AioContext *aio_context = bdrv_get_aio_context(bs);
1967 int ret;
1969 aio_context_acquire(aio_context);
1970 ret = bdrv_flush(bs);
1971 if (ret < 0 && !result) {
1972 result = ret;
1974 aio_context_release(aio_context);
1977 return result;
1981 typedef struct BdrvCoBlockStatusData {
1982 BlockDriverState *bs;
1983 BlockDriverState *base;
1984 bool want_zero;
1985 int64_t offset;
1986 int64_t bytes;
1987 int64_t *pnum;
1988 int64_t *map;
1989 BlockDriverState **file;
1990 int ret;
1991 bool done;
1992 } BdrvCoBlockStatusData;
1994 int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs,
1995 bool want_zero,
1996 int64_t offset,
1997 int64_t bytes,
1998 int64_t *pnum,
1999 int64_t *map,
2000 BlockDriverState **file)
2002 assert(bs->file && bs->file->bs);
2003 *pnum = bytes;
2004 *map = offset;
2005 *file = bs->file->bs;
2006 return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
2009 int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs,
2010 bool want_zero,
2011 int64_t offset,
2012 int64_t bytes,
2013 int64_t *pnum,
2014 int64_t *map,
2015 BlockDriverState **file)
2017 assert(bs->backing && bs->backing->bs);
2018 *pnum = bytes;
2019 *map = offset;
2020 *file = bs->backing->bs;
2021 return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
2025 * Returns the allocation status of the specified sectors.
2026 * Drivers not implementing the functionality are assumed to not support
2027 * backing files, hence all their sectors are reported as allocated.
2029 * If 'want_zero' is true, the caller is querying for mapping
2030 * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and
2031 * _ZERO where possible; otherwise, the result favors larger 'pnum',
2032 * with a focus on accurate BDRV_BLOCK_ALLOCATED.
2034 * If 'offset' is beyond the end of the disk image the return value is
2035 * BDRV_BLOCK_EOF and 'pnum' is set to 0.
2037 * 'bytes' is the max value 'pnum' should be set to. If bytes goes
2038 * beyond the end of the disk image it will be clamped; if 'pnum' is set to
2039 * the end of the image, then the returned value will include BDRV_BLOCK_EOF.
2041 * 'pnum' is set to the number of bytes (including and immediately
2042 * following the specified offset) that are easily known to be in the
2043 * same allocated/unallocated state. Note that a second call starting
2044 * at the original offset plus returned pnum may have the same status.
2045 * The returned value is non-zero on success except at end-of-file.
2047 * Returns negative errno on failure. Otherwise, if the
2048 * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are
2049 * set to the host mapping and BDS corresponding to the guest offset.
2051 static int coroutine_fn bdrv_co_block_status(BlockDriverState *bs,
2052 bool want_zero,
2053 int64_t offset, int64_t bytes,
2054 int64_t *pnum, int64_t *map,
2055 BlockDriverState **file)
2057 int64_t total_size;
2058 int64_t n; /* bytes */
2059 int ret;
2060 int64_t local_map = 0;
2061 BlockDriverState *local_file = NULL;
2062 int64_t aligned_offset, aligned_bytes;
2063 uint32_t align;
2065 assert(pnum);
2066 *pnum = 0;
2067 total_size = bdrv_getlength(bs);
2068 if (total_size < 0) {
2069 ret = total_size;
2070 goto early_out;
2073 if (offset >= total_size) {
2074 ret = BDRV_BLOCK_EOF;
2075 goto early_out;
2077 if (!bytes) {
2078 ret = 0;
2079 goto early_out;
2082 n = total_size - offset;
2083 if (n < bytes) {
2084 bytes = n;
2087 /* Must be non-NULL or bdrv_getlength() would have failed */
2088 assert(bs->drv);
2089 if (!bs->drv->bdrv_co_block_status) {
2090 *pnum = bytes;
2091 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
2092 if (offset + bytes == total_size) {
2093 ret |= BDRV_BLOCK_EOF;
2095 if (bs->drv->protocol_name) {
2096 ret |= BDRV_BLOCK_OFFSET_VALID;
2097 local_map = offset;
2098 local_file = bs;
2100 goto early_out;
2103 bdrv_inc_in_flight(bs);
2105 /* Round out to request_alignment boundaries */
2106 align = bs->bl.request_alignment;
2107 aligned_offset = QEMU_ALIGN_DOWN(offset, align);
2108 aligned_bytes = ROUND_UP(offset + bytes, align) - aligned_offset;
2110 ret = bs->drv->bdrv_co_block_status(bs, want_zero, aligned_offset,
2111 aligned_bytes, pnum, &local_map,
2112 &local_file);
2113 if (ret < 0) {
2114 *pnum = 0;
2115 goto out;
2119 * The driver's result must be a non-zero multiple of request_alignment.
2120 * Clamp pnum and adjust map to original request.
2122 assert(*pnum && QEMU_IS_ALIGNED(*pnum, align) &&
2123 align > offset - aligned_offset);
2124 *pnum -= offset - aligned_offset;
2125 if (*pnum > bytes) {
2126 *pnum = bytes;
2128 if (ret & BDRV_BLOCK_OFFSET_VALID) {
2129 local_map += offset - aligned_offset;
2132 if (ret & BDRV_BLOCK_RAW) {
2133 assert(ret & BDRV_BLOCK_OFFSET_VALID && local_file);
2134 ret = bdrv_co_block_status(local_file, want_zero, local_map,
2135 *pnum, pnum, &local_map, &local_file);
2136 goto out;
2139 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
2140 ret |= BDRV_BLOCK_ALLOCATED;
2141 } else if (want_zero) {
2142 if (bdrv_unallocated_blocks_are_zero(bs)) {
2143 ret |= BDRV_BLOCK_ZERO;
2144 } else if (bs->backing) {
2145 BlockDriverState *bs2 = bs->backing->bs;
2146 int64_t size2 = bdrv_getlength(bs2);
2148 if (size2 >= 0 && offset >= size2) {
2149 ret |= BDRV_BLOCK_ZERO;
2154 if (want_zero && local_file && local_file != bs &&
2155 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
2156 (ret & BDRV_BLOCK_OFFSET_VALID)) {
2157 int64_t file_pnum;
2158 int ret2;
2160 ret2 = bdrv_co_block_status(local_file, want_zero, local_map,
2161 *pnum, &file_pnum, NULL, NULL);
2162 if (ret2 >= 0) {
2163 /* Ignore errors. This is just providing extra information, it
2164 * is useful but not necessary.
2166 if (ret2 & BDRV_BLOCK_EOF &&
2167 (!file_pnum || ret2 & BDRV_BLOCK_ZERO)) {
2169 * It is valid for the format block driver to read
2170 * beyond the end of the underlying file's current
2171 * size; such areas read as zero.
2173 ret |= BDRV_BLOCK_ZERO;
2174 } else {
2175 /* Limit request to the range reported by the protocol driver */
2176 *pnum = file_pnum;
2177 ret |= (ret2 & BDRV_BLOCK_ZERO);
2182 out:
2183 bdrv_dec_in_flight(bs);
2184 if (ret >= 0 && offset + *pnum == total_size) {
2185 ret |= BDRV_BLOCK_EOF;
2187 early_out:
2188 if (file) {
2189 *file = local_file;
2191 if (map) {
2192 *map = local_map;
2194 return ret;
2197 static int coroutine_fn bdrv_co_block_status_above(BlockDriverState *bs,
2198 BlockDriverState *base,
2199 bool want_zero,
2200 int64_t offset,
2201 int64_t bytes,
2202 int64_t *pnum,
2203 int64_t *map,
2204 BlockDriverState **file)
2206 BlockDriverState *p;
2207 int ret = 0;
2208 bool first = true;
2210 assert(bs != base);
2211 for (p = bs; p != base; p = backing_bs(p)) {
2212 ret = bdrv_co_block_status(p, want_zero, offset, bytes, pnum, map,
2213 file);
2214 if (ret < 0) {
2215 break;
2217 if (ret & BDRV_BLOCK_ZERO && ret & BDRV_BLOCK_EOF && !first) {
2219 * Reading beyond the end of the file continues to read
2220 * zeroes, but we can only widen the result to the
2221 * unallocated length we learned from an earlier
2222 * iteration.
2224 *pnum = bytes;
2226 if (ret & (BDRV_BLOCK_ZERO | BDRV_BLOCK_DATA)) {
2227 break;
2229 /* [offset, pnum] unallocated on this layer, which could be only
2230 * the first part of [offset, bytes]. */
2231 bytes = MIN(bytes, *pnum);
2232 first = false;
2234 return ret;
2237 /* Coroutine wrapper for bdrv_block_status_above() */
2238 static void coroutine_fn bdrv_block_status_above_co_entry(void *opaque)
2240 BdrvCoBlockStatusData *data = opaque;
2242 data->ret = bdrv_co_block_status_above(data->bs, data->base,
2243 data->want_zero,
2244 data->offset, data->bytes,
2245 data->pnum, data->map, data->file);
2246 data->done = true;
2247 aio_wait_kick();
2251 * Synchronous wrapper around bdrv_co_block_status_above().
2253 * See bdrv_co_block_status_above() for details.
2255 static int bdrv_common_block_status_above(BlockDriverState *bs,
2256 BlockDriverState *base,
2257 bool want_zero, int64_t offset,
2258 int64_t bytes, int64_t *pnum,
2259 int64_t *map,
2260 BlockDriverState **file)
2262 Coroutine *co;
2263 BdrvCoBlockStatusData data = {
2264 .bs = bs,
2265 .base = base,
2266 .want_zero = want_zero,
2267 .offset = offset,
2268 .bytes = bytes,
2269 .pnum = pnum,
2270 .map = map,
2271 .file = file,
2272 .done = false,
2275 if (qemu_in_coroutine()) {
2276 /* Fast-path if already in coroutine context */
2277 bdrv_block_status_above_co_entry(&data);
2278 } else {
2279 co = qemu_coroutine_create(bdrv_block_status_above_co_entry, &data);
2280 bdrv_coroutine_enter(bs, co);
2281 BDRV_POLL_WHILE(bs, !data.done);
2283 return data.ret;
2286 int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base,
2287 int64_t offset, int64_t bytes, int64_t *pnum,
2288 int64_t *map, BlockDriverState **file)
2290 return bdrv_common_block_status_above(bs, base, true, offset, bytes,
2291 pnum, map, file);
2294 int bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes,
2295 int64_t *pnum, int64_t *map, BlockDriverState **file)
2297 return bdrv_block_status_above(bs, backing_bs(bs),
2298 offset, bytes, pnum, map, file);
2301 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t offset,
2302 int64_t bytes, int64_t *pnum)
2304 int ret;
2305 int64_t dummy;
2307 ret = bdrv_common_block_status_above(bs, backing_bs(bs), false, offset,
2308 bytes, pnum ? pnum : &dummy, NULL,
2309 NULL);
2310 if (ret < 0) {
2311 return ret;
2313 return !!(ret & BDRV_BLOCK_ALLOCATED);
2317 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2319 * Return true if (a prefix of) the given range is allocated in any image
2320 * between BASE and TOP (inclusive). BASE can be NULL to check if the given
2321 * offset is allocated in any image of the chain. Return false otherwise,
2322 * or negative errno on failure.
2324 * 'pnum' is set to the number of bytes (including and immediately
2325 * following the specified offset) that are known to be in the same
2326 * allocated/unallocated state. Note that a subsequent call starting
2327 * at 'offset + *pnum' may return the same allocation status (in other
2328 * words, the result is not necessarily the maximum possible range);
2329 * but 'pnum' will only be 0 when end of file is reached.
2332 int bdrv_is_allocated_above(BlockDriverState *top,
2333 BlockDriverState *base,
2334 int64_t offset, int64_t bytes, int64_t *pnum)
2336 BlockDriverState *intermediate;
2337 int ret;
2338 int64_t n = bytes;
2340 intermediate = top;
2341 while (intermediate && intermediate != base) {
2342 int64_t pnum_inter;
2343 int64_t size_inter;
2345 ret = bdrv_is_allocated(intermediate, offset, bytes, &pnum_inter);
2346 if (ret < 0) {
2347 return ret;
2349 if (ret) {
2350 *pnum = pnum_inter;
2351 return 1;
2354 size_inter = bdrv_getlength(intermediate);
2355 if (size_inter < 0) {
2356 return size_inter;
2358 if (n > pnum_inter &&
2359 (intermediate == top || offset + pnum_inter < size_inter)) {
2360 n = pnum_inter;
2363 intermediate = backing_bs(intermediate);
2366 *pnum = n;
2367 return 0;
2370 typedef struct BdrvVmstateCo {
2371 BlockDriverState *bs;
2372 QEMUIOVector *qiov;
2373 int64_t pos;
2374 bool is_read;
2375 int ret;
2376 } BdrvVmstateCo;
2378 static int coroutine_fn
2379 bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
2380 bool is_read)
2382 BlockDriver *drv = bs->drv;
2383 int ret = -ENOTSUP;
2385 bdrv_inc_in_flight(bs);
2387 if (!drv) {
2388 ret = -ENOMEDIUM;
2389 } else if (drv->bdrv_load_vmstate) {
2390 if (is_read) {
2391 ret = drv->bdrv_load_vmstate(bs, qiov, pos);
2392 } else {
2393 ret = drv->bdrv_save_vmstate(bs, qiov, pos);
2395 } else if (bs->file) {
2396 ret = bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read);
2399 bdrv_dec_in_flight(bs);
2400 return ret;
2403 static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque)
2405 BdrvVmstateCo *co = opaque;
2406 co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read);
2407 aio_wait_kick();
2410 static inline int
2411 bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
2412 bool is_read)
2414 if (qemu_in_coroutine()) {
2415 return bdrv_co_rw_vmstate(bs, qiov, pos, is_read);
2416 } else {
2417 BdrvVmstateCo data = {
2418 .bs = bs,
2419 .qiov = qiov,
2420 .pos = pos,
2421 .is_read = is_read,
2422 .ret = -EINPROGRESS,
2424 Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data);
2426 bdrv_coroutine_enter(bs, co);
2427 BDRV_POLL_WHILE(bs, data.ret == -EINPROGRESS);
2428 return data.ret;
2432 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2433 int64_t pos, int size)
2435 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size);
2436 int ret;
2438 ret = bdrv_writev_vmstate(bs, &qiov, pos);
2439 if (ret < 0) {
2440 return ret;
2443 return size;
2446 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
2448 return bdrv_rw_vmstate(bs, qiov, pos, false);
2451 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2452 int64_t pos, int size)
2454 QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, size);
2455 int ret;
2457 ret = bdrv_readv_vmstate(bs, &qiov, pos);
2458 if (ret < 0) {
2459 return ret;
2462 return size;
2465 int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
2467 return bdrv_rw_vmstate(bs, qiov, pos, true);
2470 /**************************************************************/
2471 /* async I/Os */
2473 void bdrv_aio_cancel(BlockAIOCB *acb)
2475 qemu_aio_ref(acb);
2476 bdrv_aio_cancel_async(acb);
2477 while (acb->refcnt > 1) {
2478 if (acb->aiocb_info->get_aio_context) {
2479 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
2480 } else if (acb->bs) {
2481 /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
2482 * assert that we're not using an I/O thread. Thread-safe
2483 * code should use bdrv_aio_cancel_async exclusively.
2485 assert(bdrv_get_aio_context(acb->bs) == qemu_get_aio_context());
2486 aio_poll(bdrv_get_aio_context(acb->bs), true);
2487 } else {
2488 abort();
2491 qemu_aio_unref(acb);
2494 /* Async version of aio cancel. The caller is not blocked if the acb implements
2495 * cancel_async, otherwise we do nothing and let the request normally complete.
2496 * In either case the completion callback must be called. */
2497 void bdrv_aio_cancel_async(BlockAIOCB *acb)
2499 if (acb->aiocb_info->cancel_async) {
2500 acb->aiocb_info->cancel_async(acb);
2504 /**************************************************************/
2505 /* Coroutine block device emulation */
2507 typedef struct FlushCo {
2508 BlockDriverState *bs;
2509 int ret;
2510 } FlushCo;
2513 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2515 FlushCo *rwco = opaque;
2517 rwco->ret = bdrv_co_flush(rwco->bs);
2518 aio_wait_kick();
2521 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2523 int current_gen;
2524 int ret = 0;
2526 bdrv_inc_in_flight(bs);
2528 if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2529 bdrv_is_sg(bs)) {
2530 goto early_exit;
2533 qemu_co_mutex_lock(&bs->reqs_lock);
2534 current_gen = atomic_read(&bs->write_gen);
2536 /* Wait until any previous flushes are completed */
2537 while (bs->active_flush_req) {
2538 qemu_co_queue_wait(&bs->flush_queue, &bs->reqs_lock);
2541 /* Flushes reach this point in nondecreasing current_gen order. */
2542 bs->active_flush_req = true;
2543 qemu_co_mutex_unlock(&bs->reqs_lock);
2545 /* Write back all layers by calling one driver function */
2546 if (bs->drv->bdrv_co_flush) {
2547 ret = bs->drv->bdrv_co_flush(bs);
2548 goto out;
2551 /* Write back cached data to the OS even with cache=unsafe */
2552 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2553 if (bs->drv->bdrv_co_flush_to_os) {
2554 ret = bs->drv->bdrv_co_flush_to_os(bs);
2555 if (ret < 0) {
2556 goto out;
2560 /* But don't actually force it to the disk with cache=unsafe */
2561 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2562 goto flush_parent;
2565 /* Check if we really need to flush anything */
2566 if (bs->flushed_gen == current_gen) {
2567 goto flush_parent;
2570 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2571 if (!bs->drv) {
2572 /* bs->drv->bdrv_co_flush() might have ejected the BDS
2573 * (even in case of apparent success) */
2574 ret = -ENOMEDIUM;
2575 goto out;
2577 if (bs->drv->bdrv_co_flush_to_disk) {
2578 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2579 } else if (bs->drv->bdrv_aio_flush) {
2580 BlockAIOCB *acb;
2581 CoroutineIOCompletion co = {
2582 .coroutine = qemu_coroutine_self(),
2585 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2586 if (acb == NULL) {
2587 ret = -EIO;
2588 } else {
2589 qemu_coroutine_yield();
2590 ret = co.ret;
2592 } else {
2594 * Some block drivers always operate in either writethrough or unsafe
2595 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2596 * know how the server works (because the behaviour is hardcoded or
2597 * depends on server-side configuration), so we can't ensure that
2598 * everything is safe on disk. Returning an error doesn't work because
2599 * that would break guests even if the server operates in writethrough
2600 * mode.
2602 * Let's hope the user knows what he's doing.
2604 ret = 0;
2607 if (ret < 0) {
2608 goto out;
2611 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2612 * in the case of cache=unsafe, so there are no useless flushes.
2614 flush_parent:
2615 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2616 out:
2617 /* Notify any pending flushes that we have completed */
2618 if (ret == 0) {
2619 bs->flushed_gen = current_gen;
2622 qemu_co_mutex_lock(&bs->reqs_lock);
2623 bs->active_flush_req = false;
2624 /* Return value is ignored - it's ok if wait queue is empty */
2625 qemu_co_queue_next(&bs->flush_queue);
2626 qemu_co_mutex_unlock(&bs->reqs_lock);
2628 early_exit:
2629 bdrv_dec_in_flight(bs);
2630 return ret;
2633 int bdrv_flush(BlockDriverState *bs)
2635 Coroutine *co;
2636 FlushCo flush_co = {
2637 .bs = bs,
2638 .ret = NOT_DONE,
2641 if (qemu_in_coroutine()) {
2642 /* Fast-path if already in coroutine context */
2643 bdrv_flush_co_entry(&flush_co);
2644 } else {
2645 co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co);
2646 bdrv_coroutine_enter(bs, co);
2647 BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE);
2650 return flush_co.ret;
2653 typedef struct DiscardCo {
2654 BdrvChild *child;
2655 int64_t offset;
2656 int bytes;
2657 int ret;
2658 } DiscardCo;
2659 static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque)
2661 DiscardCo *rwco = opaque;
2663 rwco->ret = bdrv_co_pdiscard(rwco->child, rwco->offset, rwco->bytes);
2664 aio_wait_kick();
2667 int coroutine_fn bdrv_co_pdiscard(BdrvChild *child, int64_t offset, int bytes)
2669 BdrvTrackedRequest req;
2670 int max_pdiscard, ret;
2671 int head, tail, align;
2672 BlockDriverState *bs = child->bs;
2674 if (!bs || !bs->drv) {
2675 return -ENOMEDIUM;
2678 if (bdrv_has_readonly_bitmaps(bs)) {
2679 return -EPERM;
2682 ret = bdrv_check_byte_request(bs, offset, bytes);
2683 if (ret < 0) {
2684 return ret;
2687 /* Do nothing if disabled. */
2688 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2689 return 0;
2692 if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) {
2693 return 0;
2696 /* Discard is advisory, but some devices track and coalesce
2697 * unaligned requests, so we must pass everything down rather than
2698 * round here. Still, most devices will just silently ignore
2699 * unaligned requests (by returning -ENOTSUP), so we must fragment
2700 * the request accordingly. */
2701 align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment);
2702 assert(align % bs->bl.request_alignment == 0);
2703 head = offset % align;
2704 tail = (offset + bytes) % align;
2706 bdrv_inc_in_flight(bs);
2707 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_DISCARD);
2709 ret = bdrv_co_write_req_prepare(child, offset, bytes, &req, 0);
2710 if (ret < 0) {
2711 goto out;
2714 max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX),
2715 align);
2716 assert(max_pdiscard >= bs->bl.request_alignment);
2718 while (bytes > 0) {
2719 int num = bytes;
2721 if (head) {
2722 /* Make small requests to get to alignment boundaries. */
2723 num = MIN(bytes, align - head);
2724 if (!QEMU_IS_ALIGNED(num, bs->bl.request_alignment)) {
2725 num %= bs->bl.request_alignment;
2727 head = (head + num) % align;
2728 assert(num < max_pdiscard);
2729 } else if (tail) {
2730 if (num > align) {
2731 /* Shorten the request to the last aligned cluster. */
2732 num -= tail;
2733 } else if (!QEMU_IS_ALIGNED(tail, bs->bl.request_alignment) &&
2734 tail > bs->bl.request_alignment) {
2735 tail %= bs->bl.request_alignment;
2736 num -= tail;
2739 /* limit request size */
2740 if (num > max_pdiscard) {
2741 num = max_pdiscard;
2744 if (!bs->drv) {
2745 ret = -ENOMEDIUM;
2746 goto out;
2748 if (bs->drv->bdrv_co_pdiscard) {
2749 ret = bs->drv->bdrv_co_pdiscard(bs, offset, num);
2750 } else {
2751 BlockAIOCB *acb;
2752 CoroutineIOCompletion co = {
2753 .coroutine = qemu_coroutine_self(),
2756 acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num,
2757 bdrv_co_io_em_complete, &co);
2758 if (acb == NULL) {
2759 ret = -EIO;
2760 goto out;
2761 } else {
2762 qemu_coroutine_yield();
2763 ret = co.ret;
2766 if (ret && ret != -ENOTSUP) {
2767 goto out;
2770 offset += num;
2771 bytes -= num;
2773 ret = 0;
2774 out:
2775 bdrv_co_write_req_finish(child, req.offset, req.bytes, &req, ret);
2776 tracked_request_end(&req);
2777 bdrv_dec_in_flight(bs);
2778 return ret;
2781 int bdrv_pdiscard(BdrvChild *child, int64_t offset, int bytes)
2783 Coroutine *co;
2784 DiscardCo rwco = {
2785 .child = child,
2786 .offset = offset,
2787 .bytes = bytes,
2788 .ret = NOT_DONE,
2791 if (qemu_in_coroutine()) {
2792 /* Fast-path if already in coroutine context */
2793 bdrv_pdiscard_co_entry(&rwco);
2794 } else {
2795 co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco);
2796 bdrv_coroutine_enter(child->bs, co);
2797 BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE);
2800 return rwco.ret;
2803 int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf)
2805 BlockDriver *drv = bs->drv;
2806 CoroutineIOCompletion co = {
2807 .coroutine = qemu_coroutine_self(),
2809 BlockAIOCB *acb;
2811 bdrv_inc_in_flight(bs);
2812 if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) {
2813 co.ret = -ENOTSUP;
2814 goto out;
2817 if (drv->bdrv_co_ioctl) {
2818 co.ret = drv->bdrv_co_ioctl(bs, req, buf);
2819 } else {
2820 acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
2821 if (!acb) {
2822 co.ret = -ENOTSUP;
2823 goto out;
2825 qemu_coroutine_yield();
2827 out:
2828 bdrv_dec_in_flight(bs);
2829 return co.ret;
2832 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2834 return qemu_memalign(bdrv_opt_mem_align(bs), size);
2837 void *qemu_blockalign0(BlockDriverState *bs, size_t size)
2839 return memset(qemu_blockalign(bs, size), 0, size);
2842 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
2844 size_t align = bdrv_opt_mem_align(bs);
2846 /* Ensure that NULL is never returned on success */
2847 assert(align > 0);
2848 if (size == 0) {
2849 size = align;
2852 return qemu_try_memalign(align, size);
2855 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
2857 void *mem = qemu_try_blockalign(bs, size);
2859 if (mem) {
2860 memset(mem, 0, size);
2863 return mem;
2867 * Check if all memory in this vector is sector aligned.
2869 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2871 int i;
2872 size_t alignment = bdrv_min_mem_align(bs);
2874 for (i = 0; i < qiov->niov; i++) {
2875 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2876 return false;
2878 if (qiov->iov[i].iov_len % alignment) {
2879 return false;
2883 return true;
2886 void bdrv_add_before_write_notifier(BlockDriverState *bs,
2887 NotifierWithReturn *notifier)
2889 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
2892 void bdrv_io_plug(BlockDriverState *bs)
2894 BdrvChild *child;
2896 QLIST_FOREACH(child, &bs->children, next) {
2897 bdrv_io_plug(child->bs);
2900 if (atomic_fetch_inc(&bs->io_plugged) == 0) {
2901 BlockDriver *drv = bs->drv;
2902 if (drv && drv->bdrv_io_plug) {
2903 drv->bdrv_io_plug(bs);
2908 void bdrv_io_unplug(BlockDriverState *bs)
2910 BdrvChild *child;
2912 assert(bs->io_plugged);
2913 if (atomic_fetch_dec(&bs->io_plugged) == 1) {
2914 BlockDriver *drv = bs->drv;
2915 if (drv && drv->bdrv_io_unplug) {
2916 drv->bdrv_io_unplug(bs);
2920 QLIST_FOREACH(child, &bs->children, next) {
2921 bdrv_io_unplug(child->bs);
2925 void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size)
2927 BdrvChild *child;
2929 if (bs->drv && bs->drv->bdrv_register_buf) {
2930 bs->drv->bdrv_register_buf(bs, host, size);
2932 QLIST_FOREACH(child, &bs->children, next) {
2933 bdrv_register_buf(child->bs, host, size);
2937 void bdrv_unregister_buf(BlockDriverState *bs, void *host)
2939 BdrvChild *child;
2941 if (bs->drv && bs->drv->bdrv_unregister_buf) {
2942 bs->drv->bdrv_unregister_buf(bs, host);
2944 QLIST_FOREACH(child, &bs->children, next) {
2945 bdrv_unregister_buf(child->bs, host);
2949 static int coroutine_fn bdrv_co_copy_range_internal(
2950 BdrvChild *src, uint64_t src_offset, BdrvChild *dst,
2951 uint64_t dst_offset, uint64_t bytes,
2952 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags,
2953 bool recurse_src)
2955 BdrvTrackedRequest req;
2956 int ret;
2958 /* TODO We can support BDRV_REQ_NO_FALLBACK here */
2959 assert(!(read_flags & BDRV_REQ_NO_FALLBACK));
2960 assert(!(write_flags & BDRV_REQ_NO_FALLBACK));
2962 if (!dst || !dst->bs) {
2963 return -ENOMEDIUM;
2965 ret = bdrv_check_byte_request(dst->bs, dst_offset, bytes);
2966 if (ret) {
2967 return ret;
2969 if (write_flags & BDRV_REQ_ZERO_WRITE) {
2970 return bdrv_co_pwrite_zeroes(dst, dst_offset, bytes, write_flags);
2973 if (!src || !src->bs) {
2974 return -ENOMEDIUM;
2976 ret = bdrv_check_byte_request(src->bs, src_offset, bytes);
2977 if (ret) {
2978 return ret;
2981 if (!src->bs->drv->bdrv_co_copy_range_from
2982 || !dst->bs->drv->bdrv_co_copy_range_to
2983 || src->bs->encrypted || dst->bs->encrypted) {
2984 return -ENOTSUP;
2987 if (recurse_src) {
2988 bdrv_inc_in_flight(src->bs);
2989 tracked_request_begin(&req, src->bs, src_offset, bytes,
2990 BDRV_TRACKED_READ);
2992 /* BDRV_REQ_SERIALISING is only for write operation */
2993 assert(!(read_flags & BDRV_REQ_SERIALISING));
2994 if (!(read_flags & BDRV_REQ_NO_SERIALISING)) {
2995 wait_serialising_requests(&req);
2998 ret = src->bs->drv->bdrv_co_copy_range_from(src->bs,
2999 src, src_offset,
3000 dst, dst_offset,
3001 bytes,
3002 read_flags, write_flags);
3004 tracked_request_end(&req);
3005 bdrv_dec_in_flight(src->bs);
3006 } else {
3007 bdrv_inc_in_flight(dst->bs);
3008 tracked_request_begin(&req, dst->bs, dst_offset, bytes,
3009 BDRV_TRACKED_WRITE);
3010 ret = bdrv_co_write_req_prepare(dst, dst_offset, bytes, &req,
3011 write_flags);
3012 if (!ret) {
3013 ret = dst->bs->drv->bdrv_co_copy_range_to(dst->bs,
3014 src, src_offset,
3015 dst, dst_offset,
3016 bytes,
3017 read_flags, write_flags);
3019 bdrv_co_write_req_finish(dst, dst_offset, bytes, &req, ret);
3020 tracked_request_end(&req);
3021 bdrv_dec_in_flight(dst->bs);
3024 return ret;
3027 /* Copy range from @src to @dst.
3029 * See the comment of bdrv_co_copy_range for the parameter and return value
3030 * semantics. */
3031 int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, uint64_t src_offset,
3032 BdrvChild *dst, uint64_t dst_offset,
3033 uint64_t bytes,
3034 BdrvRequestFlags read_flags,
3035 BdrvRequestFlags write_flags)
3037 trace_bdrv_co_copy_range_from(src, src_offset, dst, dst_offset, bytes,
3038 read_flags, write_flags);
3039 return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
3040 bytes, read_flags, write_flags, true);
3043 /* Copy range from @src to @dst.
3045 * See the comment of bdrv_co_copy_range for the parameter and return value
3046 * semantics. */
3047 int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, uint64_t src_offset,
3048 BdrvChild *dst, uint64_t dst_offset,
3049 uint64_t bytes,
3050 BdrvRequestFlags read_flags,
3051 BdrvRequestFlags write_flags)
3053 trace_bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
3054 read_flags, write_flags);
3055 return bdrv_co_copy_range_internal(src, src_offset, dst, dst_offset,
3056 bytes, read_flags, write_flags, false);
3059 int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset,
3060 BdrvChild *dst, uint64_t dst_offset,
3061 uint64_t bytes, BdrvRequestFlags read_flags,
3062 BdrvRequestFlags write_flags)
3064 return bdrv_co_copy_range_from(src, src_offset,
3065 dst, dst_offset,
3066 bytes, read_flags, write_flags);
3069 static void bdrv_parent_cb_resize(BlockDriverState *bs)
3071 BdrvChild *c;
3072 QLIST_FOREACH(c, &bs->parents, next_parent) {
3073 if (c->role->resize) {
3074 c->role->resize(c);
3080 * Truncate file to 'offset' bytes (needed only for file protocols)
3082 int coroutine_fn bdrv_co_truncate(BdrvChild *child, int64_t offset,
3083 PreallocMode prealloc, Error **errp)
3085 BlockDriverState *bs = child->bs;
3086 BlockDriver *drv = bs->drv;
3087 BdrvTrackedRequest req;
3088 int64_t old_size, new_bytes;
3089 int ret;
3092 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3093 if (!drv) {
3094 error_setg(errp, "No medium inserted");
3095 return -ENOMEDIUM;
3097 if (offset < 0) {
3098 error_setg(errp, "Image size cannot be negative");
3099 return -EINVAL;
3102 old_size = bdrv_getlength(bs);
3103 if (old_size < 0) {
3104 error_setg_errno(errp, -old_size, "Failed to get old image size");
3105 return old_size;
3108 if (offset > old_size) {
3109 new_bytes = offset - old_size;
3110 } else {
3111 new_bytes = 0;
3114 bdrv_inc_in_flight(bs);
3115 tracked_request_begin(&req, bs, offset - new_bytes, new_bytes,
3116 BDRV_TRACKED_TRUNCATE);
3118 /* If we are growing the image and potentially using preallocation for the
3119 * new area, we need to make sure that no write requests are made to it
3120 * concurrently or they might be overwritten by preallocation. */
3121 if (new_bytes) {
3122 mark_request_serialising(&req, 1);
3124 if (bs->read_only) {
3125 error_setg(errp, "Image is read-only");
3126 ret = -EACCES;
3127 goto out;
3129 ret = bdrv_co_write_req_prepare(child, offset - new_bytes, new_bytes, &req,
3131 if (ret < 0) {
3132 error_setg_errno(errp, -ret,
3133 "Failed to prepare request for truncation");
3134 goto out;
3137 if (!drv->bdrv_co_truncate) {
3138 if (bs->file && drv->is_filter) {
3139 ret = bdrv_co_truncate(bs->file, offset, prealloc, errp);
3140 goto out;
3142 error_setg(errp, "Image format driver does not support resize");
3143 ret = -ENOTSUP;
3144 goto out;
3147 ret = drv->bdrv_co_truncate(bs, offset, prealloc, errp);
3148 if (ret < 0) {
3149 goto out;
3151 ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3152 if (ret < 0) {
3153 error_setg_errno(errp, -ret, "Could not refresh total sector count");
3154 } else {
3155 offset = bs->total_sectors * BDRV_SECTOR_SIZE;
3157 /* It's possible that truncation succeeded but refresh_total_sectors
3158 * failed, but the latter doesn't affect how we should finish the request.
3159 * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */
3160 bdrv_co_write_req_finish(child, offset - new_bytes, new_bytes, &req, 0);
3162 out:
3163 tracked_request_end(&req);
3164 bdrv_dec_in_flight(bs);
3166 return ret;
3169 typedef struct TruncateCo {
3170 BdrvChild *child;
3171 int64_t offset;
3172 PreallocMode prealloc;
3173 Error **errp;
3174 int ret;
3175 } TruncateCo;
3177 static void coroutine_fn bdrv_truncate_co_entry(void *opaque)
3179 TruncateCo *tco = opaque;
3180 tco->ret = bdrv_co_truncate(tco->child, tco->offset, tco->prealloc,
3181 tco->errp);
3182 aio_wait_kick();
3185 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
3186 Error **errp)
3188 Coroutine *co;
3189 TruncateCo tco = {
3190 .child = child,
3191 .offset = offset,
3192 .prealloc = prealloc,
3193 .errp = errp,
3194 .ret = NOT_DONE,
3197 if (qemu_in_coroutine()) {
3198 /* Fast-path if already in coroutine context */
3199 bdrv_truncate_co_entry(&tco);
3200 } else {
3201 co = qemu_coroutine_create(bdrv_truncate_co_entry, &tco);
3202 bdrv_coroutine_enter(child->bs, co);
3203 BDRV_POLL_WHILE(child->bs, tco.ret == NOT_DONE);
3206 return tco.ret;